]> git.ipfire.org Git - thirdparty/Python/cpython.git/commit
gh-115398: Revert `PyExpat_CAPI_MAGIC` version bump (GH-116411)
authorSebastian Pipping <sebastian@pipping.org>
Wed, 6 Mar 2024 17:55:07 +0000 (18:55 +0100)
committerGitHub <noreply@github.com>
Wed, 6 Mar 2024 17:55:07 +0000 (09:55 -0800)
commit8a8e9204d1f737fa5dc6f5087411bdbe8139d3b2
treec5105fa0fd1a16d25034aa836fcf55a15cb1004d
parentc012c8ab7bb72a733bd98be5df32e262b9045f1a
gh-115398: Revert `PyExpat_CAPI_MAGIC` version bump (GH-116411)

Revert "gh-115398: Increment PyExpat_CAPI_MAGIC for SetReparseDeferralEnabled addition (GH-116301)"

This reverts part of commit eda2963378a3c292cf6bb202bb00e94e46ee6d90.  Why? this comment buried in an earlier code review explains:

I checked again how that value is used in practice, it's here:

https://github.com/python/cpython/blob/0c80da4c14d904a367968955544dd6ae58c8101c/Modules/_elementtree.c#L4363-L4372

Based on that code my understanding is that loading bigger structs from the future is considered okay unless `PyExpat_CAPI_MAGIC` differs, which implies that (1) magic needs to stay the same to support loading the future from the past and (2) that `PyExpat_CAPI_MAGIC` should only ever change for changes that do not increase size (but keep it constant).

To summarize, that supports your argument.
I checked branches 3.8, 3.9, 3.10, 3.11, 3.12 now and they all have the same comparison code there so reverting that magic string bump will support seamless backporting.
Include/pyexpat.h